Anthropic's advanced AI model, Claude Mythos Preview, successfully developed novel attacks against two highly studied cryptographic systems: a weakened version of AES and HAWK, a post-quantum digital signature scheme. Initially, Claude stated it was impossible to improve known attacks on AES, but after persistent prompting from researchers, it discovered a technique called Möbius Bridge.

This technique can accelerate the best comparable attack against an experimental seven-round version of AES by 200 to 800 times, though it doesn't break the full AES-128 used widely today due to extreme conditions required for the attack. The more significant breakthrough was against HAWK-256, a candidate for post-quantum standardization by NIST. Claude identified a previously unexploited mathematical symmetry, drastically reducing the effort needed to recover a key from 2⁶⁴ to 2³⁸ operations for its smaller variant. While this doesn't destroy post-quantum cryptography, it highlights a weakness that would require increasing key sizes for HAWK.

The true importance of these findings lies not in breaking current internet security, but in demonstrating that an AI can now compete with human experts in complex cybersecurity fields like cryptography. The AES attack cost about 100,000 dollars and took Claude a week, while human researchers needed a month to verify it. This shows AI's capability to reason about mathematical properties of algorithms and produce novel techniques, shifting who can participate in the ongoing race between cryptographic defenders and attackers.

👉 Read the full article on Gizmodo en Español

Microsoft is preparing to launch a "super app" for its AI assistant Copilot, as announced by CEO Satya Nadella. This super app aims to integrate various Copilot experiences, including chat, coworking tools, coding assistance, and autonomous agents, spanning both consumer and commercial use cases. The initiative, internally dubbed "Delivering one Copilot," seeks to unify the product's personal and enterprise functionalities, allowing users to switch between them.

The concept of AI super apps, which combine chatbots and AI agents, is a growing trend in the AI industry, with companies like OpenAI (hinting at a ChatGPT super app) and Anthropic (with Claude Cowork) also moving in this direction. Microsoft's move reflects the industry's belief that AI chatbots are evolving towards more comprehensive, integrated platforms.

This announcement comes as Microsoft reported stronger-than-expected revenue growth, driven by its cloud business, and increased engagement with Copilot. Nadella noted that conversations per user nearly doubled year-over-year, and weekly engagement is on par with Outlook and Teams, indicating rapid user adoption.

👉 Read the full article on Gizmodo

The content of this article was inaccessible via the provided URL. Therefore, a deep summary cannot be generated.

Based on the title, "Microsoft is openly competing with OpenAI, Anthropic more than ever," the article likely discusses the increasing competitive landscape in the AI industry, particularly between Microsoft and its partners/rivals OpenAI and Anthropic. This competition could manifest in areas such as LLM development, AI agent capabilities, and the deployment of Generative AI solutions.

👉 Read the full article on TechCrunch